Introduction
(Article introduction authored by INCA Editorial Team)
Sepsis is a life-threatening condition that frequently occurs in ICU patients with major trauma due to their compromised immune systems and severe injuries.
The prevalence of sepsis in this patient population ranges from 30.4% to 33.3%, leading to high mortality rates (19.5%-26.7%) and significant healthcare costs.
Risk factors include age, abnormal vital signs, comorbidities, injury severity, immunosuppression, and invasive procedures, but their complex interactions limit predictive accuracy.
While clinical prediction models have been developed, they lack universal applicability and external validation. Traditional statistical models have struggled to capture these complexities, necessitating advanced predictive solutions.
Artificial intelligence (AI) has emerged as a powerful tool in medical prediction, offering advantages such as rapid data processing, improved clinical decision-making, and early disease detection.
AI-based models can analyze vast datasets to provide precise and personalized predictions. This study aims to develop and validate a machine-learning-based AI application to assess the risk of sepsis in ICU patients with major trauma.
Methods
This study involved 961 patients, with a prospective analysis of data from 244 patients with major trauma at our hospital and a retrospective analysis of data from 717 patients extracted from a database in the United States.
The patients in the model development cohort were randomly divided into a training cohort and an internal validation cohort at a ratio of 8:2.
Various machine-learning algorithms, including logistic regression, decision tree, extreme gradient boosting machine (eXGBM), neural network (NN), random forest & light gradient boosting machine (LightGBM), were employed.
Results
The incidence of sepsis in the model development cohort was 43.44%. Twelve predictors—gender, abdominal trauma, open trauma, red blood cell count, heart rate, respiratory rate, injury severity score (ISS), sequential organ failure assessment (SOFA) score, Glasgow coma scale (GCS), smoking, total protein levels, and hematocrit—were used as input variables.
Internal validation revealed that the NN model achieved the highest area under the curve (AUC) of 0.932, followed by LightGBM (0.913) and eXGBM (0.912).
In external validation, eXGBM (AUC: 0.891) performed best, followed by LightGBM (0.886), whereas the NN model had a lower AUC of 0.787. Considering both validation cohorts, LightGBM emerged as the most optimal model, scoring 82 points in predictive performance and was subsequently deployed as an online AI application.
This study successfully developed and validated an AI-based mobile application to predict sepsis risk in ICU patients with major trauma.
By integrating 12 predictive variables, the AI platform demonstrated strong reliability and generalizability. The model’s external validation further reinforced its applicability to diverse clinical settings.
This AI application can enhance clinical decision-making by enabling early identification of high-risk patients, facilitating timely interventions, and potentially reducing sepsis-related morbidity and mortality.
Conclusion
Sepsis remains a leading cause of ICU mortality, and trauma patients are particularly vulnerable. Our findings align with previous research identifying gender, ISS, GCS, heart rate, albumin, respiratory rate, and transfusion history as sepsis predictors. This study also highlights the significance of abdominal trauma, open trauma, and smoking as additional risk factors.
Open wounds and abdominal injuries increase infection risks due to bacterial exposure, while smoking weakens immune responses and respiratory function, elevating susceptibility to pneumonia and sepsis.
The integration of AI in sepsis prediction represents a transformative approach to critical care. Machine-learning models like LightGBM offer enhanced accuracy and adaptability compared to traditional statistical methods.
By leveraging AI-driven predictive models, healthcare professionals can improve patient outcomes, optimize resource allocation, and mitigate the substantial healthcare burden associated with sepsis in ICU trauma patients.
Future research should focus on refining AI models, incorporating additional biomarkers, and conducting further external validation to ensure widespread clinical applicability.
Source: Sun, Baisheng MDa,b; Lei, Mingxing MDa,d,f; Wang, Li MDb; Wang, Xiaoli MMb; Li, Xiaoming MMg; Mao, Zhi MDb; Kang, Hongjun MD, PhDb; Liu, Hui MDb; Sun, Shiying MDc; Zhou, Feihu MD, PhDb,e,*. Prediction of sepsis among patients with major trauma using artificial intelligence: a multicenter validated cohort study. International Journal of Surgery 111(1):p 467-480, January 2025. DOI: 10.1097/JS9.0000000000001866
